Since the previous combined af and MCP update, AI assistants can work in many more AmpliFlow workflows. The largest change concerns projects: assistants can now follow activity, prioritize the portfolio, use templates, manage the project lifecycle, and copy tasks between projects with clear safety boundaries.

This update covers published af releases through 6.16.0 and hosted MCP improvements deployed through 22 August 2026.

What’s new

Projects and tasks

af now uses AmpliFlow as the only source for deciding which tasks are ready. Assignment, start date, order, completion, and the blocked tag control the queue without a separate local task state.
AI assistants can show project activity, mark their own events as read, favorite projects, and retrieve portfolio, timeline, and workspace summaries.
Assistants can publish project status updates, request a fresh update, and see unanswered requests. These narratives remain separate from the project’s manual status.
Assistants can preview AmpliFlow project templates and create a complete draft or launched project with tasks, subtasks, milestones, and calculated dates.
Projects can be listed separately as active, archived, or deleted. Archive, restore, return-to-draft, and trash recovery now use explicit, verified actions.
AI assistants can copy a project task to another project without changing the source. The copy keeps the task fields, assignees, subtasks, comments, attachments, tags, and milestones that AmpliFlow can recreate, and the result reports both preserved and unsupported parts.
Task cloning reads the full source description and confirms that the instructions exist on the destination before copying anything else. If the description cannot be verified, af removes the incomplete copy and leaves the source untouched.
Project files now use the project’s Pages structure. Existing commands for creating a folder, listing, uploading, downloading, and deleting files use the same document surface as AmpliFlow.

Checklists

AI assistants can now distinguish reusable checklist templates, checklists that people actually carry out, and recurring checklists that create future runs.

af checklist-template manages templates, af checklist manages planned and completed checklists, and af recurring-checklist manages recurrences. MCP uses the same concepts.
Lists and detail views return the actual checklist, its status, dates, steps, activities, stored values, and source template. Readonly MCP sessions can inspect the information but cannot start or change a checklist.
AI assistants can create reusable steps with several ordered task lists. Each list can have its own guidance plus required and optional items, and empty lists are rejected before anything is saved.
Recurring checklists can be created monthly or quarterly, on either a fixed date or a rule such as the second Tuesday of the month.
A user without a saved time zone can start a checklist by supplying a valid IANA time zone for that run only.

Process maps and risks

Assistants can read AmpliFlow’s process-mapping guidance before suggesting changes.
af and MCP can create published core processes with purpose, inputs, outputs, identified risks, targets, and responsible people. The returned process identity is verified before the next action.
Assistants can create subprocesses and process steps with activities and risk links, update existing steps, and assign responsible users or teams.
Process maps can be given positions, sizes, colors, connections, labels, and decision points. Assistants can also save drafts with visible layout changes before changing the published map.
MCP can manage the map’s Input and Output system nodes, connect upstream and downstream processes, and highlight linked risks. Risk lists per process step show the exact relationship, even when the same risk is linked to several steps.

Organization, impact, and business registers

Administrators can build users, roles, teams, competencies, positions, and competency matrices in dependency order through af or MCP. Passwords are accepted only through protected input and never appear in plain text.
AI assistants can configure bilingual impact categories with five score levels before grading goals, projects, risks, and improvements. Readonly assistants can inspect the categories but cannot change them.
af and MCP can work with more native AmpliFlow registers: customer industries, customers, item types and units, items, equipment, purchase orders, KPIs, and year wheels.
News can be targeted to users and teams or published directly at a chosen time. Supplier contacts can be updated, and improvement history shows every value saved in each activity.

Time, calendars, and follow-up

Project time can be recorded without linking it to one task. AmpliFlow checks that the time lands on the right row in the timesheet, even when the same user has already recorded time on the same project and day.
ICS calendars can be connected to time tracking, validated, and given rules that match calendar events to the right work. Calendar URLs are treated as secrets and are never printed.
Assistants can run filtered timesheet reports, export the selection to Excel, and find completed weeks that have not been submitted.
Improvements can be analyzed by template, impact, status, and impacted process. Results can be drilled down to individual improvements and exported.

Pages and local file work

The experimental Linux command af page mount exposes Pages, folders, and stored files as a local filesystem. Text pages can be edited, while PDFs, images, and other stored files can be added, read, moved, and deleted within AmpliFlow’s file limits.
Concurrent local and remote text edits are preserved as separate versions instead of silently overwriting one another.

Improvements

Connection and authorization

The hosted MCP connection now stays responsive even after many reconnects from ChatGPT, Claude, Copilot, and other clients.
An MCP client can reconnect after a restart without registering again.
Administrators can review and record their organization’s acceptance of AmpliFlow’s AI terms before enabling the AI feature.
Stored af identities can be refreshed in unattended workflows without the password appearing on the command line.
In af-managed Git workflows, commits keep the user’s own author details.

Safer automation

MCP readonly mode keeps read tools visible but hides write tools for checklists, projects, processes, organization records, registers, and other writable areas.
Project-scoped commands reject a task from another project instead of risking a write to the wrong record.
Custom-list file fields can upload, download, and preserve one or several files in the same format used by the web interface.

Fixed bugs

Project task comments can again be listed, created, edited, and deleted after AmpliFlow removed its older comment routes.
Publishing and metadata changes for checklist templates preserve the selected steps. Rerunning the publication creates no duplicate versions.
Custom-list files once again appear in the AmpliFlow interface, several files in one field remain intact, and archived list items can be deleted correctly.
Year-wheel items are created with the required defaults expected by AmpliFlow.
Organization lists show a clear error message when records cannot be saved with durable references, instead of showing a partial list.

Want to try the new workflows? Update af or reconnect your MCP client, then ask the assistant to read the current information before it proposes a change.

This page is also available in Swedish.